’ volume 15, no. 2 When one door is closed, don’t you know, another is open. Imprint january 2015 — Bob Marley 1. Introduction1 In its simplest form, the principle states that, neces- sarily, if you know that P, and you know that if P then Q, then you also know that Q. Or, at least, you would then be in a position to easily An Open and Shut Case: know Q by simple inference. Modifications to the principle have been proposed in response to various problems, both obvious and techni- cal. But these problems won’t concern us here, so we’ll stick with the simple formulation. Epistemic Closure in the The epistemic closure principle lies at the crossroads of many trends in the theory of . Many tout it as a linchpin in the most compelling skeptical (e. g. Stroud 1984). Some rely on it to motivate contextualist hypotheses about the semantics of knowledge Manifest Image attributions (DeRose 1995, 2009; Cohen 1999). And others appeal to it in the course of motivating views about the relationship between knowledge and practical interests (e. g. Hawthorne 2004). If epistemic closure fails, then the landscape of contemporary will need to be reconfigured.

1.1 Two arguments against closure There are two main arguments against the epistemic closure principle. The first is that rejecting the closure principle is “the only John Turri way to avoid ” (Dretske 2013: 32, emphasis in the original). Suppose that George is standing in front of an audience waving his University of Waterloo hands around while rehearsing a provocative (Moore 1959: ch. 7; see also Stine 1976). The skeptic argues:

1. If George knows that he’s standing up, then he knows that he’s not merely dreaming that he’s standing up.

© 2015 John Turri 2. But George doesn’t know that he’s not merely dreaming This work is licensed under a Creative Commons that he’s standing up. Attribution-NonCommercial-NoDerivatives 3.0 License. 1. I dedicate this paper to the memory of , whose generosity and insight were truly inspiring.

john turri An Open and Shut Case

3. So George doesn’t know that he’s standing up. epistemic closure prevents skepticism from corroding our knowledge of important day-to-day matters: “We can have most of our cake (the Some anti-skeptics respond: only part worth eating) and eat it too” (Dretske 2013: 33). 1. If George knows that he’s standing up, then he knows that Although this is a way to respond to the skeptic, it is certainly not he’s not merely dreaming that he’s standing up. the only way to do so. For instance, contextualists have argued that their own response to skepticism is better because it not only lets us 2*. And George does know that he’s standing up. count many ordinary knowledge attributions as true, but it also allows 3*. So George knows that he’s not merely dreaming that he’s us to retain the epistemic closure principle (DeRose 1995; Lewis 1996; standing up. Cohen 2013). And there are non-contextualist responses that diagnose the skeptic’s mistake in terms of a subtle modal fallacy (Sosa 1999) or a Despite ending up with radically different conclusions, our skeptic pragmatic confusion about illocutionary force or the norms of speech and anti-skeptic start from the same place, namely, by accepting that if acts (Turri 2010a; Turri 2012; Turri 2014a). George knows that he’s standing up, then he knows that he’s not mere- There is a better argument against the epistemic closure principle ly dreaming. A shared commitment to the epistemic closure principle (traceable in some form to Dretske 1970; see Turri 2014b: §19). If a explains the common starting point of the two arguments. The skeptic is closed under known entailment, then the and anti-skeptic then make different use of that shared conditional following conditional is true: if you have the attitude toward P, and premise. The skeptic relies on the that George can’t know you know that P entails Q, then you also have that same proposi- that he isn’t just dreaming and then concludes that George doesn’t tional attitude toward Q. I regret that I attended a play last night. I know that he’s standing up; the anti-skeptic relies on the intuition that also know that if I attended a play last night, then I existed last night. George knows that he’s standing up and then concludes that George From these two facts does it follow that I regret that I existed last does know that he isn’t merely dreaming. night? Far from it. Regret isn’t closed under known entailment. The Dretske (1970) and Nozick (1981) advocate a dramatically different same is true for other propositional attitudes. Gilmore might be hap- response to the skeptic. They accept that George doesn’t know that he py that his sister’s broken arm is healing quickly, and Gilmore knows isn’t merely dreaming (line 2 of the skeptic’s argument). At the same that if her broken arm is healing quickly, then her arm was broken, time, they accept that George knows that he’s standing up (line 2* of but he’s not happy that her arm was broken. I might be surprised (or the anti-skeptic’s argument). Dretske and Nozick reconcile both of angry, or sad) that James won the competition, and I know that if those things by rejecting line 1 of each argument. That is, they deny James won, then someone won, but I’m not surprised (or angry, or that if George knows that he’s standing up, then George knows that he sad) that someone won. Similar remarks apply to other mental states, isn’t merely dreaming. such as perceiving or remembering. By giving up the epistemic closure principle, we “grant the skeptic Setting knowledge aside so that we don’t beg the question, there a partial victory” (Nozick 1981: 270). Some think that the victory is well isn’t a single propositional attitude that is closed under known entail- earned, because it strikes us as a “suspicious” exercise in “bad faith” ment. (On certain formulations of closure, might be an exception to claim that we know that we’re not merely dreaming (Nozick 1981: to this generalization.) But knowledge is a propositional attitude. So, 264). Nevertheless, the victory is only partial, because the failure of absent special considerations to the contrary, it’s likely that knowledge

philosophers’ imprint – 2 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

isn’t closed under known entailment either. Without a special reason violates the epistemic closure principle (2004: 38). Theoretical pro- to do so, we should not treat knowledge differently from its kindred. posals to abandon closure are “revisionary” and suffer from “deep and wide-ranging” “conflict” with our “intuition” (Hawthorne 2004: 36). 1.2 Two arguments for closure In light of all this, closure’s proponents find it “striking” (Hawthorne That brings us to the two main arguments in favor of closure. First, 2004: 36) and “startling” (Fumerton 1987) that “some philosophers closure’s proponents point out that competent deduction from known have gone so far as” to deny the principle and, in the process, “bla- premises is “in general a way of coming to” know one’s conclusion, of tantly violate” such a “basic and extremely plausible intuition” (Steup “extending one’s knowledge” (Williamson 2000: 117; Hawthorne 2004: 2005). They marvel at the “absurdity” of such a proposal — though they 33; Stanley 2005: 18–20, 94). The epistemic closure principle is then do admit that it’s less absurd than “denying the transitivity of the taller glossed as a way of “articulating” this “extremely intuitive ” about than relation” (Hawthorne 2004: 38; compare Hawthorne 2013: 45). competent deduction (Hawthorne 2004: 33). But this train of thought Related to these claims about ordinary practice and the absurdity is far too hasty. Nothing so strong as a necessarily true generalization of denying closure in the abstract, philosophers have also claimed that is needed to vindicate the general epistemic efficacy of competent de- conjunctive assertions which suggest a violation of closure are “abom- duction. Saving and investing prudently is in general a way of retiring inable” (DeRose 1995). Such conjunctions include ‘She knows that she comfortably, but there are of course exceptions (compare Turri 2010b). has hands, but she doesn’t know that she isn’t a handless brain-in- It seems likely that there is a more modest explanation for the efficacy a-vat’ and ‘He knows that his car is parked in the lot, but he doesn’t of deduction. We should rule out more modest explanations before know that his car hasn’t been stolen.’ reaching for a necessarily true generalization. This rather lavish gloss on ordinary practice and what “we” find Second, and much more commonly, closure’s proponents rely heav- intuitively compelling or conversationally “abominable” has thus ily on the claim that epistemic closure is a defining feature of ordi- far gone unchallenged, even by closure’s opponents (Nozick 1981: nary thought and talk about knowledge — that is, of folk epistemology. 205–206; Dretske 2013: 32). Thus it is widely assumed that closure’s As puts it, epistemic closure is “just a familiar fact about opponents bear the revisionist’s burden of either explaining away human knowledge, something we all recognize and abide by in our contrary or showing that their view delivers otherwise un- thought and talk about knowing things”: “anyone who speaks about attainable benefits. knowledge and understands what others say about it will recognize” this (Stroud 1984: 18–19). (Actually, Stroud seems to have in a 1.3 The state of the art and the way forward principle that is logically stronger than the epistemic closure principle The “only anti-skeptical game in town” argument against closure is un- as characterized above.) Stewart Cohen claims that the principle de- derwhelming, as is the suggestion that closure is the only or best way serves to be called “axiomatic” (Cohen 1999: 69), and he defends his to vindicate the epistemological efficacy of competent deduction. The closure-based version of by appealing to “our intuitions” best reason to reject the epistemic closure principle is the inductive and the dispositions and judgments of “competent speakers” (Cohen argument from the fact that no other propositional attitude is closed 2013: 69, 71, et passim; see also Feldman 1999; Williamson 2000; Bon- under known entailment. Unless there is evidence that knowledge Jour 2010). John Hawthorne claims that it’s “very difficult to put one- is somehow different, it’s likely that knowledge is not closed under self in a frame of mind” where one attributes knowledge in a way that known entailment either. The best reason to think that knowledge is

philosophers’ imprint – 3 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

somehow different is that epistemic closure is a defining feature of different cover story.) I also use two versions of the story that varied folk epistemology and that abandoning it is repugnant to intuition and the source of belief. In addition to the long history in philosophy of em- . phasizing differences among the sources of knowledge (Priest 2007; Participants on all sides of the debate accept that patterns in ordi- Sgarbi 2012), this decision was motivated by recent research dem- nary thought and speech are a valuable source of information which onstrating that source of belief affects knowledge attribution (Turri place vital constraints on theories in this area. For example, Jonathan 2014c; Friedman & Turri 2014). The two sources I chose to work with Vogel aims for a theory of knowledge and epistemic closure that can are and inference. I will return to the significance of differ- “accommodate the body of our intuitions in an unforced, convinc- entiating sources in the Conclusion. ing way” (1990: 298). In laying out the “ground rules” by which we I should also like to emphasize that I will ultimately distinguish judge competing theories in this area, Hawthorne appeals repeatedly between general or unqualified versions of the closure principle, on the to what he views as “uncontroversial” and “widespread” “inclinations” one hand, and source-relative versions of the principle, on the other. to ascribe or deny knowledge in certain cases — that is, “our intuitive The philosophical debate has thus far focused on the unqualified ver- verdicts” (Hawthorne 2004: 7, 8, 12, et passim; compare Cohen 2013: sion, but, as will become clear, the evidence warrants a more fine- 71). And Dretske proposes that “ordinary, intuitive judgments” about grained approach. a “set of clear cases” can be “used to test” competing hypotheses (1981: 92–93). Dretske even goes so far as to report results from his “own poll 1.4 Two reasons to value the method among nonphilosophers” (Dretske 1981: 249, n. 8). Some philosophers will immediately object, “When it comes to philo- Although the claim that closure is a core principle of folk episte- sophical inquiry, it doesn’t matter what ordinary people think. So this mology has gone unchallenged thus far, no serious evidence been empirical investigation is worthless.” This dismissive reaction is mis- produced in its favor. It is a straightforwardly empirical claim that or- guided, for two reasons. dinary practice exhibits such patterns. This claim has thus far been First, for anyone even passingly familiar with the literature on clo- granted based on anecdotal evidence. But much better evidence is sure over the past three decades, it’s utterly obvious that empirical facts obtainable, so it is unwise to rest content with mere anecdote. At this about ordinary practice have mattered greatly to serious philosophical point, the best way forward is to get better evidence. debate on this topic. I am simply investigating rigorously what others In this paper, I investigate whether epistemic closure really is a have assumed to be true based on personal impressions and anecdote. core commitment of folk epistemology — that is, to paraphrase Stroud, Second, and setting aside the particulars of recent philosophical whether the closure principle is “embodied” in our ordinary ways of theorizing about closure specifically, it’s doubtful that ordinary think- thinking and talking about knowledge. If it is thus embodied, then the ing is irrelevant to philosophical theorizing, for two reasons. case for closure is strengthened. If it isn’t, then the case for closure is On the one hand, philosophers are people too. It’s naive to sup- seriously weakened and the inductive argument against closure pre- pose that initiation into the vocation of philosophy insulates us from vails, at least for now. the habits and tendencies of the human condition. As In order to conduct this investigation, I’ll rely primarily on a “car once wrote, philosophers may “derive much of [their] sense of the theft case” adapted from Vogel’s influential work on epistemic closure. whole from the pre-reflective orientation which is our common heri- (Towards the end of the paper, I replicate the basic findings using a tage” (Sellars 1963: 3). If some of those habits and tendencies lead us

philosophers’ imprint – 4 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

astray, then we should want to know whether they’re influencing our we may “know [our] way around in it reflectively with no intellectual theoretical reflections. A particularly striking example of this is the holds barred”. recent discovery of a human tendency called excuse validation (Turri Sellars viewed “the great speculative systems of ancient and me- 2013a; Turri & Blouw 2014). This is the surprising tendency to incor- dieval philosophy”, “the major schools of contemporary Continental rectly view someone who blamelessly breaks a rule as having not broken thought”, and “common sense” and “ordinary usage” philosophy as the rule at all. When asked to evaluate a situation in which someone aiming for “more or less adequate accounts of the manifest image”. Sel- definitely breaks a rule but does so blamelessly, roughly half of people lars also thought that recent “analytic” philosophy, especially as influ- agree that a rule was broken, but equally many people disagree that a enced by “the later Wittgenstein”, had done “increasing justice to the rule was broken. Those who disagree aren’t being disingenuous or in- manifest image”, having made progress toward “isolating it in some- sincere. Rather, their desire to excuse leads them to judge that no rule thing like its pure form”. has been broken. (This is roughly the inverse phenomenon of blame Although the manifest image shapes and influences “philosophi- validation, the well-documented tendency whereby a desire to blame cal thinking itself”, the image has “an existence which transcends . . . distorts factual judgments; see Alicke 1992; Alicke & Rose 2010.) the individual thought of individual thinkers”. And because the image It’s surprising that people are so prone to misdescribing such cases. “transcends the individual thinker, there is and error with respect to What’s more, even in their most reflective moments, professional phi- it”. Despite our intimate relationship with the manifest image, we can losophers exhibit the same tendency. Excuse validation occurs in the and do get it wrong. highest reaches of contemporary philosophical debates. For instance, Sellars sums this all up nicely in the following memorable passage: the most common objection to the knowledge account of assertion Much of academic philosophy can be interpreted as an perfectly fits the mold of excuse validation (Turri 2013a). attempt by individual thinkers to delineate the manifest On the other hand, some of the most illustrious and accomplished image (not recognized, needless to say, as such) — an philosophers have argued that distilling the defining features of ordinary image which is both immanent in and transcendent of thought is an essential part of the ’s job. For example, ac- their thinking. In this respect, a philosophy can be evalu- cording to Sellars, philosophy aims to “understand how things in the ated as perceptive or imperceptive, mistaken or correct, broadest possible sense of the term hang together in the broadest even though one is prepared to say that the image [it] possible sense of the term” — that is, to find in all things “unity at the delineate[s] is but one way in which reality appears to the reflective level” (all Sellars quotes are from Sellars 1963: ch. 1, sec- human mind. And it is, indeed, a task of the first impor- tions 2–4). As part of this, the philosopher is tasked with reconciling tance to delineate this image. “the manifest image” — our ordinary ways of the world and our place in it, “sophisticated common sense”, our “pre-reflective In light of this, I submit that Sellars would have strongly approved of orientation” and “heritage” — with the deliverances of scientific in- the present investigation and, moreover, that much of the recent trend quiry or “the scientific image”. In order to do this, philosophy has to known broadly as “experimental philosophy” coheres seamlessly with get the manifest image right, which is “a task of the first importance”. the Sellarsian vision of philosophy’s proper scope and method. Philosophy aims to “understand the structure of this image”, so that

philosophers’ imprint – 5 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

2. Experiment 1: Open and closed between-subjects After reading the story, participants were instructed, “Please read the statements below and check all that are true in the story you just 2.1 Method read.” Each participant was presented with three statements. Partici- Participants (N = 145)2 were randomly assigned to one of four con- pants in all conditions received statements [2] and [3]; participants in ditions in a 2 (Source: Perception, Inference) × 2 (Content: Positive, Positive conditions also received statement [1P]; participants in Nega- Negative) between-subjects design. Participants in each condition tive conditions also received statement [1N]. (I use numerical labels read a single story. The Source factor manipulated source of belief. for expository convenience; participants never saw the labels.) The Content factor manipulated which of two statements participants [1P] Maxwell knows that his car is parked in C8. were asked (Positive: the person knows where the car is parked; Nega- tive: the person knows that the car has not been stolen). Here is the [1N] Maxwell knows that his car has not been stolen. story, with the difference between Perception and Inference condi- [2] Maxwell is in the archives room. tions bracketed and separated by a slash: [3] Maxwell is in his assistant’s office. (Perception/Inference) When Mr. Maxwell arrives at work in the morning, he always parks in one of two spots: The order of the statements was rotated randomly. C8 or D8. Half the time he parks in C8, and half the time I will interpret participants in Positive conditions as ascribing to he parks in D8. Today Maxwell parked in C8. ¶3 It’s lunch- Maxwell knowledge that the car is parked in C8 if and only if they se- time at work. Maxwell and his assistant are up in the ar- lect option [1P]. Similarly I will interpret participants in Negative con- chives room searching for a particular document. Max- ditions as ascribing to Maxwell knowledge that the car has not been well says, “I might have left the document in my car.” The stolen if and only if they select option [1N]. For convenience, I will use assistant asks, “Mr. Maxwell, is your car parked in space the following terminology: C8? It’s not unheard of for cars to be stolen.” ¶ Maxwell • People who selected option [1P] ascribed positive knowledge. [looks carefully out the window / thinks carefully for a moment] and then responds, “No, my car has not been • People who selected option [1N] ascribed negative knowledge. stolen. It is parked in C8.” Nine participants who either omitted option [2] or selected option [3] were excluded from the , on the grounds that they were not reading the story carefully enough. Including data from these partici- 2. Sixty-three female, aged 18–69 years, mean age = 29.7 years; 92% reporting pants didn’t affect the results reported below. English as a native language. Participants were recruited and tested using an online platform (Amazon Mechanical Turk + Qualtrics) and compensated $0.30 for approximately 2 minutes of their time. Participation was restricted 2.2 Results and discussion to United States residents. Participants were not allowed to retake the sur- If epistemic closure is a principle of folk epistemology, then partici- vey; repeat participation was prevented by screening AMT Worker IDs. They pants should not ascribe positive knowledge significantly more fre- filled out a brief demographic survey after testing. quently than they ascribe negative knowledge. By contrast, if they do 3. Indicates paragraph break on the participant’s screen.

philosophers’ imprint – 6 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

These results are strong initial evidence that epistemic closure is not a principle of folk epistemology. However, it could be objected that because participants in the Positive and Negative conditions were presented with different sets of statements, they interpreted those statements in relevantly different ways. In response to this objection, first, I note that if the positive and negative knowledge statements themselves precipitated a relevant change in context, then we would expect that to happen in both Inference and Perception conditions. But there is no evidence that this happened to participants in Perception conditions. So the objection seems unlikely to be correct. Second, and setting aside the first response entirely, the next section reports an ex- periment that avoids the objection by simultaneously presenting both knowledge statements to all participants. Fig. 1. Experiment 1. Percent of participants attributing knowledge across four conditions. 3. Experiment 2: Open and (almost) closed within-subjects A between-subjects design is potentially open to the worry that the ques- tions themselves precipitate a relevant change of context. Perhaps just ascribe positive knowledge more frequently, then that is evidence that asking about car theft discourages knowledge attribution in a way that epistemic closure is not a principle of folk epistemology. mentioning car parking does not. If that’s right, then a between-subjects The results were absolutely clear: participants ascribed positive design can’t control for the change of context caused by the difference knowledge significantly more frequently (Figure 1). But higher posi- in questions asked. Given the amount of attention paid to contextualist tive attribution occurred only in the Inference condition (92% / 42%, epistemology over the past two decades, we shouldn’t ignore this possi- Fisher’s exact test, p < .00001, Cramer’s V = .528, all tests two-tailed). bility (Unger 1984; Cohen 1988; DeRose 1995; Lewis 1996). Accordingly, In the Perception condition, rates of negative attribution were slightly in this experiment I adopt a within-subjects design in which all the par- higher, though the difference was not significant (95% / 88%, Fisher’s ticipants are asked about both knowledge attributions. exact test, p = .411, n.s.). Rate of knowledge attribution in the Positive Inference condition didn’t differ significantly from rates in either the 3.1 Method 5 Positive Perception (Fisher’s exact test, p = .699) or the Negative Per- Participants (N = 123) were randomly assigned to one of two condi- ception conditions (Fisher’s exact test, p = 1). It was the Negative Infer- tions, Perception and Inference. The stories and procedures were ex- ence condition that differed from the other three.4 content of Maxwell’s belief from positive to negative, the odds of denying knowledge increased over 37 times more for inferential belief than for percep- 4. In a logistic regression model that included Source, Content, and the interac- tual belief. See Appendix 1. (Compare the results reported in Turri 2014c.) tion term Source*Content as predictors and knowledge denial as the outcome, only the interaction Source*Content significantly predicted knowledge de- 5. Thirty-six female, aged 18–73 years, mean age = 30.5 years; 95% reporting nial, recording an odds ratio of over 37. This means that when changing the English as a native language.

philosophers’ imprint – 7 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

actly the same as in Experiment 1, except for one important difference: Again the results were absolutely clear. We observe the same basic each participant was presented with both the positive and the nega- pattern as last time: positive knowledge attribution was significant- tive knowledge statements (i. e. 1P and 1N from Experiment 1). I ex- ly higher (Figure 2). But this time higher knowledge attribution oc- cluded data from 5 participants who failed comprehension questions. curred in both the Inference condition (81% / 36%, binomial test, p < Including data from these participants doesn’t change the pattern of .000001) and the Perception condition (95% / 80%, binomial test, p < results reported below. .001). Still, the difference between positive and negative attribution was much larger in the Inference condition. Rates of knowledge at- 3.2 Results and discussion tribution were significantly above what could be expected by chance If the objection considered against Experiment 1 has merit, then in for positive attribution in Perception (binomial test, p < .000001), for this experiment positive knowledge attribution should not be signifi- negative attribution in Perception (binomial test, p = .000001), and cantly higher than negative knowledge attribution. By contrast, if posi- for positive attribution in Inference (binomial test, p = .000002). By tive knowledge attribution is once again significantly higher, then that contrast, the rate of negative knowledge attribution in Inference was provides strong evidence that epistemic closure is not a principle of significantly below chance (binomial test, p = .048). folk epistemology. It is also worth examining subgroups in our sample. In particular, it is worth examining the number of people who opted for the four possible combinations of response to the two knowledge statements (Figure 2). In the Perception condition, the most common pattern was to ascribe both positive and negative knowledge (52 of 65, 80%). By contrast, in the Inference condition, the most common pattern was to ascribe positive knowledge and deny negative knowledge (26 of 58, 45%) — that is, the pattern which conflicts directly with the epistemic closure principle. The number of participants in the Infer- ence condition who selected this combination exceeded what could be expected by chance (binomial test, p < .001, test proportion = .25), whereas the number who selected it in the Perception condition was trending below what could be expected by chance (10 of 68, 15%, binomial test, p = .09, test proportion = .25). Interestingly, in the en- Fig. 2. Experiment 2. Left panel: percent of participants tire study, not a single participant ascribed negative knowledge and attributing knowledge across two source conditions with denied positive knowledge. positive/negative attribution as a within-subjects factor. 4. Experiment 3: The myth of the abominable conjunction Right panel: number of participants in the two source conditions opting for the four possible combinations of Many philosophers follow Keith DeRose in claiming that conjunc- knowledge attribution. P = positive knowledge attribu- tive assertions that suggest a violation of closure are “abominable” tion; N = negative knowledge attribution. (DeRose 1995: 27–29; see also Schaffer 2007; Ichikawa & Steup 2012:

philosophers’ imprint – 8 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

section 5.1). Such conjunctions include ‘She knows that she has hands, [3] Maxwell knows that his car is parked in C8. But he does not but she doesn’t know that she isn’t a handless brain-in-a-vat’ and ‘He know that his car has not been stolen. knows that his car is still parked in the lot, but he doesn’t know that [4] Maxwell does not know that his car is parked in C8. And he his car hasn’t been stolen.’ These have also been called “ridiculous” does not know that his car has not been stolen. (Dretske 2013: 31) and “distinctly repugnant” (Sosa 2004: 41). Accord- ingly, it might be alleged that we should not trust the results from Ex- The order of the four options was rotated randomly. periment 2, because participants weren’t forced to confront just how The second difference was that comprehension checks were not unacceptably counterintuitive it sounds to agree that Maxwell knows included this time, for two reasons: First, in the first two experiments, that his car is parked in the lot but deny that Maxwell knows that his including those who had failed a comprehension check didn’t change car hasn’t been stolen. Perhaps participants successfully “compart- the results. Second, and more importantly, the conjunctions them- mentalized” their two choices, to the point where many opted for a selves are long and complicated, and I didn’t want to increase the cog- closure-denying combination without fully grasping what they had nitive load much beyond that of the first two experiments. done. The present experiment tests this conjecture by forcing partici- pants to explicitly choose among a set of conjunctions. 4.2 Results and discussion If the objection considered against Experiment 2 has merit, then in this 4.1 Method experiment very few if any participants should choose the closure-de- Participants (N = 80)6 were randomly assigned to one of two condi- nying combination [3]. By contrast, if many participants choose com- tions, Perception and Inference. The stories and procedures were ex- bination [3], then that further undermines the claim that epistemic actly the same as in Experiment 1, except for two important differenc- closure is a core commitment of folk epistemology. es. The first difference was that this time participants were instructed, Yet again the results were absolutely clear and cohered extremely “Please read the pairs of statements below and select the pair that best well with our earlier observations (Figure 3). There was a significant describes Maxwell in the story you just read.” Each participant was effect of condition on which conjunction participants selected as best presented with four conjunctions, corresponding to the four possible (χ2(1, N = 80) = 27, p < .00001, Cramer’s V = .581). In the Perception combinations of answers to the positive and negative knowledge attri- condition, the vast majority chose option [1] (34 of 40, 80%) and very butions: Positive & Negative, ~Positive & Negative, Positive & ~Nega- few chose the closure-denying option [3] (3 of 40, 7.5%). By contrast, tive, ~Positive & ~Negative. in the Inference condition, participants chose option [1] at rates no different from what could be expected by chance (32.5%, binomial [1] Maxwell knows that his car is parked in C8. And he knows test, p = .358, test proportion = .25), and a shockingly large majority that his car has not been stolen. chose the closure-denying option [3] (25 of 40, 63%), which far ex- [2] Maxwell does not know that his car is parked in C8. But he ceeds what could be expected by chance (binomial test, p < .000001, does know that his car has not been stolen. test proportion = .25) — indeed, significantly more participants chose option [3] than chose the other three options combined (binomial test,

6. Thirty-one female, aged 18–70 years, mean age = 29.2 years; 93% reporting p < .000001, test proportion = .35). English as a native language. These results provide further powerful evidence that ordinary practice is not unqualifiedly committed to epistemic closure. They

philosophers’ imprint – 9 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

that Maxwell’s negative belief is improperly based, not because they implicitly reject closure. The present experiment addresses this con- cern by making Maxwell’s final statement more clearly express an in- ference. Second, some participants who accept closure might none- theless think that Maxwell fails to know a relevant conditional, such as if his car has been stolen, then it isn’t parked in C8. For such participants, attributing positive knowledge while denying negative knowledge shouldn’t count as evidence against closure, because closure requires that a relevant conditional be known. The present experiment ad- dresses this concern by asking participants to also consider a relevant conditional knowledge attribution.

5.1 Method Fig. 3. Experiment 3. Number of participants in the two Participants (N = 153)8 were randomly assigned to one of two condi- Source conditions who chose the different conjunctions tions, Perception and Inference. The stories and procedures were very as best describing the case. P = positive knowledge at- similar to those used in Experiment 2, except for three differences. tribution; N = negative knowledge attribution. First, this time the story ends with Maxwell saying, “My car is parked in C8, so it has not been stolen.” This addresses the concern that the basis of Maxwell’s negative belief is unclear. Second, participants were instructed slightly differently: “At the end of the story, which things should also make us seriously reconsider whether the “abominable” did Maxwell know? Please check all that apply.”9 Third, all participants conjunctions really are intolerable or unacceptably counterintuitive. were presented with a list of five candidate pieces of knowledge: Is the abominable conjunction any less mythical than the abominable snowman? [1] His car is parked in C8.

5. Experiment 4: Basis clarified, conditional included [2] His car has not been stolen. At least two concerns about the earlier studies might be raised at this [3] If his car has been stolen, then it’s not parked in C8. point.7 First, perhaps it wasn’t clear enough to participants that Max- [4] He is in the archives room. well believed that his car hasn’t been stolen based on his belief (or the fact) that his car is parked in C8. This could cause us to misinterpret some responses because some participants who accept closure could 8. Fifty-eight female, aged 18–67 years, mean age = 30.9 years; 95% reporting English as a native language. still attribute positive knowledge but deny negative knowledge. In 9. Given the complexity of the conditional knowledge attribution (i. e. [3] on the particular, they might opt for that combination because they suspect list of five statements), it seemed desirable to eliminate the “Maxwell knows” from each of the separate options and instead include it in the instructions for 7. Thanks to and Peter Klein for discussion on these points. evaluating the options.

philosophers’ imprint – 10 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

[5] He is in his assistant’s office.

The addition of [3] enables us to address the concern that some partic- ipants were being misinterpreted as implicitly rejecting closure. More specifically, it enables us to analyze responses from only those partici- pants who explicitly attributed relevant conditional knowledge. Options [4] and [5] were included to keep cognitive load closely matched across experiments. Participants checked [4] and omitted [5] 91% of the time. In light of this very high rate, the fact that in earlier studies excluding those who failed comprehension checks made no difference, and the fact that knowledge attributions arguably should not count as comprehension checks, I did not exclude any participants Fig. 4. Experiment 4. Left panel: overall percent of partici- from the analysis. Nevertheless, excluding participants who gave the pants attributing knowledge across the two Source conditions “wrong” answer to [4] or [5] does not change the pattern of results re- with Positive/Negative attribution as a within-subjects factor. ported below. Right panel: including data from only those participants who attributed relevant conditional knowledge. 5.2 Results and discussion The overall results were similar to those from Experiment 2 (Figure 4), knowledge at rates below chance (40%, binomial, p = .015).10 Includ- except that this time higher positive knowledge attribution occurred ing data from only participants who attributed conditional knowledge, only in the Inference condition (86% / 36%, binomial test, p < .000001). we’re left with respectable sample sizes of 31 for Inference and 30 for In the Perception condition, positive attribution was not significantly Perception. Once again, in this more restricted sample, positive attribu- higher than negative attribution (92% / 86%, binomial test, p = .084). tion was significantly higher than negative attribution in the Inference Positive knowledge attribution was significantly above chance in Per- condition (61% / 23%, binomial test, p = .00001), whereas positive at- ception and Inference conditions (binomial, ps < .000001), as was negative knowledge attribution in Perception (binomial, p < .000001). 10. This might come as a surprise to some readers, though perhaps it shouldn’t, because the conditional is more complex and effortful to comprehend. In By contrast, negative attribution in Inference was significantly below the present case, this is compounded by two further facts: (i) the condi- chance (binomial, p = .015). Once again, negative inferential belief was tional has an affirmative antecedent and a negative consequent, which can inhibit fluent processing, and (ii) the conditional is embedded in a mental viewed differently from the other three categories. state attribution, which further complicates matters. Furthermore, the prag- In order to fully address the concerns outlined earlier, it’s impor- matics of conditionals might have played a further role in this. For instance, as anyone who has taught introductory logic knows, people sometimes in- tant to examine the data specifically from participants who attributed terpret a conditional as asserting or implying both the antecedent and the conditional knowledge to Maxwell (i. e. selected option [3]) (Figure 4). consequent — i. e. the conjunctive interpretation. The vast majority of partici- There was no effect of condition on whether participants attributed pants (87%) said Maxwell knows that his car is parked in C8, (binomial, p < .000001, test proportion = .5). At least some of those participants might conditional knowledge (Perception/Inference: 39% / 41%, Fisher’s have interpreted the conditional conjunctively, in which case they would exact test, p = .870) and overall participants attributed conditional naturally deny that Maxwell knows the conditional, because they think the “right conjunct” (i. e. the consequent) is false.

philosophers’ imprint – 11 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

tribution was not significantly higher in the Perception condition (97% doubt that this suggestion is correct, it’s nevertheless true that it would / 90%, binomial, p = .368). Positive and negative attribution exceeded strengthen the main findings if we observed the same basic pattern chance in the Perception condition (binomials, ps < .000001). In the using a different cover story. The present experiment pursues this inference condition, negative knowledge attribution was significantly question. Having observed the same basic patterns over the first four below chance (binomial, p = .003), whereas positive attribution was experiments, I’ll limit the replication to the basic between-subjects de- non-significantly above chance (binomial, p = .281). sign from Experiment 1. Examining subgroups in our sample again reveals a picture very similar to that revealed by Experiment 2. In the Perception condi- 6.1 Method tion, the most common pattern by far was to ascribe both positive Participants (N = 160)11 were randomly assigned to one of four con- and negative knowledge (62 of 77, 80%), which exceeds what could ditions in a 2 (Source: Perception, Inference) × 2 (Content: Positive, be expected by chance (binomial test, p < .000001, test proportion = Negative) between-subjects design, which exactly mirrored the design .25). By contrast, in the Inference condition, the most common pattern from Experiment 1. Participants in each condition read a single story. was to ascribe positive knowledge and deny negative knowledge (36 This time the story was about an office worker who sets her computer of 76, 47%), which exceeds what could be expected by chance (bino- to work and then walks across the hall into her assistant’s office. The mial, p = .00001, test proportion = .25). Focusing specifically on only Source factor manipulated source of belief. The Content factor manip- those who attributed conditional knowledge, the same combinations ulated which of two statements participants were given (Positive: the were again most frequently selected in both Perception (27 of 30, computer is analyzing data; Negative: hackers have not crashed the 90%) and Inference (12 of 31, 39%). Whereas in Experiment 2 not a computer). Here is the story, with the difference between Perception single participant ascribed negative and denied positive knowledge, and Inference conditions bracketed and separated by a slash: this time a very small fraction of participants selected this combina- (Perception/Inference) When Mrs. Palmer arrives at tion (5 of 153, 3%). work in the morning, she always starts her computer 6. Experiment 5: Conceptual replication and puts it to work analyzing data sets. Today she put her computer to work analyzing demographic data and The results thus far have painted a nuanced and consistent picture that then walked into her assistant’s office across the hall. ¶ seriously undermines the persistent claims of epistemic closure’s pro- Palmer says to her assistant, “It might take the computer ponents. But all the studies used the same basic “car theft” story. This is a while to analyze the demographic data.” The assistant good in some respects. For example, given that we used the same story asks, “Mrs. Palmer, is your computer analyzing the data? in both Experiment 1 and Experiment 2, if we found a closure-violating It’s not unheard of for computer hackers to attack the asymmetry in the first but not the second experiment, then we could system and crash a computer.” ¶ Mrs. Palmer [looks care- have attributed that to the difference between a between-subjects and fully through the window into her office / thinks care- a within-subjects comparison. And closure’s proponents might have fully for a moment about the matter] and then responds, found some solace in that asymmetry. However, using the same story

is not good in all respects. For instance, it might be suggested that 11. Fifty-six female, aged 18–62 years, mean age = 29.5 years; 96% reporting Eng- all the results are somehow an artifact of this particular story. While I lish as a native language.

philosophers’ imprint – 12 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

“No, the hackers have not crashed it. The computer is analyzing data.”

This “computer crash” story closely matches the earlier “car theft” story in terms of length and complexity. After reading the story, participants were instructed, “Please read the statements below and check all that are true in the story you just read.” Each participant was presented with three statements. Partici- pants in all conditions received statements [2] and [3]; participants in Positive conditions also received statement [1P]; participants in Nega- tive conditions also received statement [1N].

[1P] Palmer knows that the computer is analyzing data.

[1N] Palmer knows that hackers have not crashed the computer. Fig. 5. Experiment 5. Percent of participants attributing knowledge. [2] Palmer is in the archives room.

[3] Palmer is in her assistant’s office.

The order of the statements was rotated randomly. I will interpret par- test, p = .481, n.s.). Rate of knowledge attribution in the Positive Infer- ticipant response to the knowledge statement the same way as in Ex- ence condition didn’t differ significantly from rates in either the Posi- periments 1 and 2 (i. e. in terms of attributing positive and negative tive Perception (Fisher’s exact test, p = .765) or the Negative Perception knowledge). conditions (Fisher’s exact test, p = .331). Once again it was the Negative Seventeen participants who either omitted option [3] or selected Inference condition that differed from the other three.12 option [2] were excluded from the analysis, on the grounds that they 7. Conclusion: Experiment as stereoscope were not reading the story carefully enough. Including data from these participants didn’t affect the results reported below. For decades debate has raged over the epistemic closure principle. On one side, Dretske, Nozick and others argue that there are good theo- 6.2 Results and discussion retical arguments for rejecting the epistemic closure principle and that The results replicated our earlier findings: participants ascribed posi- there are compelling counterexamples to it. On the other side, Stroud,

tive knowledge significantly more frequently (Figure 5). But, again, 12. In a logistic regression model that included Source, Content, and the interac- higher positive attribution occurred only in the Inference condition tion term Source*Content as predictors and knowledge denial as the outcome, (81% / 50%, Fisher’s exact test, p = .005, Cramer’s V = .320). In the only the interaction Source*Content significantly predicted knowledge deni- al, recording an odds ratio of 7.5. This means that when changing the content Perception condition, rates of negative attribution were slightly higher, of Maxwell’s belief from positive to negative, the odds of denying knowledge though the difference was not significant (92% / 87%, Fisher’s exact increased 7.5 times more for inferential belief than for perceptual belief. See Appendix B. (Compare the results reported in Turri 2014c.)

philosophers’ imprint – 13 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

Vogel, Hawthorne, and others argue that the epistemic closure prin- feature of folk epistemology. This is ironic given that the most famous ciple is a defining feature of ordinary thought and talk about knowl- attempted counterexamples feature perceptually based beliefs (e. g. edge. Some have gone so far as to claim that denying closure is only Dretske’s zebra case). slightly less absurd than “denying the transitivity of the taller than rela- Despite the irony, these results fit well with the results from recent tion” (Hawthorne 2004: 38). research on knowledge attributions. Studies have repeatedly shown But the results from our five experiments paint a very different im- that when the agent has, or has had, perceptual access to the truth of age of the status of epistemic closure in folk epistemology. The overall what she , it significantly increases the rate at which people pattern of results is definitely not what closure’s proponents had led us ascribe knowledge to her, even in cases which philosophers claim to expect. We repeatedly observed patterns that are hard to reconcile are obviously not instances of knowledge, such as Gettier cases (Star- with the claim that (unqualified) epistemic closure is a defining feature mans & Friedman 2012; Turri 2013b; Nagel, San Juan, & Mar 2013; of folk epistemology. We observed similar results using between-sub- Turri, Buckwalter, & Blouw 2014; for an overview, see Turri in press). jects and within-subjects designs, when questioning participants in I suspect that this growing body of research points to something very different ways, and when using different cover stories. Participants re- important about the way we ordinarily think of experience and its re- sponded in the same basic way even when it required them to provide lationship to knowledge — about perception’s epistemic profile in the purportedly “abominable” or “repugnant” answers. But this happened manifest image. It’s probably no accident that Hume’s discussion of primarily only when the protagonist relied on inference rather than “skeptical doubts” began only with matters of fact that go “beyond the observation, a pattern observed in all five experiments. In Experiment present testimony of our senses” (Hume 1748: §4.1), or that Moore’s fa- 3, participants overwhelmingly selected closure-friendly conjunctions mous proof featured readily perceptible extremities (Moore 1959: 144), in the perception condition, but a surprisingly large majority in the or that even 3-year-olds assign perceptual access considerable weight inference condition selected the supposedly “abominable” closure- when deciding whether other people know (Pillow 1989). denying conjunction. In Experiments 1 and 5, negative attribution was Our results have implications for the many debates in contempo- slightly but non-significantly higher in the perception condition; in rary epistemology where proponents of epistemic closure wrap them- Experiment 4, positive attribution was slightly but non-significantly selves in the mantle of common sense and ordinary practice. These higher; and in Experiment 2, positive attribution was higher. Overall, include debates over skepticism and the semantics and pragmatics of it’s possible that these slight differences in positive/negative attribu- knowledge attributions. If the unqualified (i. e. non-source-relative) tions in perception conditions resulted from random variation, and epistemic closure principle is not a defining feature of folk epistemol- the population is disposed to attribute positive and negative percep- ogy, then theories that “abandon” it are not revisionary. Rejecting the tual knowledge at basically the same rate. However, the same cannot principle doesn’t conflict with ordinary practice or common sense. be said for the positive/negative asymmetry in inference conditions. In light of our results, accusations of abandonment and revisionism Overall, then, ordinary practice does not endorse an unqualified ring decidedly hollow. Instead, a theory of knowledge that rejects the version of the epistemic closure principle. But it’s arguably consistent unqualified principle thereby gains the advantage of being consistent with our results that ordinary practice is committed to a source-relative with ordinary practice on this score, and perhaps even the consider- version of the epistemic closure principle. In particular, it’s consistent able advantage of being able to charitably explain our overall pattern with our results that a perceptual epistemic closure principle is a defining of knowledge judgments.

philosophers’ imprint – 14 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

One explanation of our results is that what philosophers mean by some of which always close when you pass through, others of which ‘epistemic closure’ — philosophers’ closure — is too subtle or complicated often remain open. for ordinary people to understand, which is why they exhibit patterns Finally, our results also suggest an explanation for the persistent that violate it.13 I have four responses to this explanation. First, and controversy over the epistemic closure principle in contemporary most importantly, if it’s correct, then the principle is admittedly not a epistemology. Sellars suggested that not only does philosophy aim for defining feature of folk epistemology and, by extension, abandoning an accurate picture of the manifest image, but the image is both im- it is not revisionary, in which case a serious re-evaluation of several manent in and transcendent of individual philosophical reflection, si- core debates in contemporary epistemology is in order. Second, the multaneously infusing it and at least partly eluding it. If some doors to explanation might be unduly pessimistic about the sophistication of knowledge are open and some closed, and this fact is reflected in our folk epistemology. For instance, ordinary people are remarkably sensi- “pre-reflective orientation” and “heritage” (i. e. part of folk epistemol- tive to the complex factors that undermine knowledge in Gettier cases ogy), then it’s possible that each side of the debate had grasped part of (Turri, Buckwalter, & Blouw 2014), which are at least as subtle and the truth while failing to grasp that its grasp is only partial. Each side complicated as the epistemic closure principle. Third, the explanation intuitively glimpsed part of the image (the image immanent) but mis- doesn’t account for why people seem to respect closure in perceptual took it for the whole (the image transcendent). If that explanation is cases but not in inferential cases. Fourth, if people responded con- on the right track, then our present investigation has handsomely re- sistently with closure across all the sources of knowledge, then that paid the effort. Through the use of careful and theoretically informed would support the view that they implicitly understand and accept an experimentation, we have taken “two differing perspectives on a land- unqualified closure principle. By the same token, if people respond scape” and, in good stereoscopic fashion, “fused them into one coher- consistently with closure for some but not all sources of knowledge, ent experience” (Sellars 1963: 4–5). then that supports the view that people implicitly understand and ac- cept closure for some sources but not for others. Acknowledgments Of course, there might still be good theoretical arguments favor- ing the unqualified closure principle. Rather than relying on what are, For helpful feedback and discussion, I thank James Beebe, Matt Benton, at the very least, problematic descriptive claims about ordinary prac- Peter Blouw, Wesley Buckwalter, Fred Dretske, Ori Friedman, Alvin tice, closure’s proponents could pursue a prescriptive strategy: make Goldman, Peter Klein, Joshua Knobe, Blake Myers, David Rose, Ernest the argument that we ought to think of knowledge that way. To con- Sosa, Angelo Turri, and Jonathan Vogel. Thanks also to audiences at tinue with a Sellarsian thought quoted earlier: closure’s proponents Rutgers University and the Buffalo Experimental Philosophy Confer- can rightfully point out that “there is truth and error with respect to ence. This research was supported by the Social Sciences and Humani- the [manifest image], even though the image itself might have to be rejected, ties Research Council of Canada and an Early Researcher Award from in the last analysis, as false” (Sellars 1963: 14, emphasis in the original). the Ontario Ministry of Economic Development and Innovation. Alternatively, they could adopt a more nuanced approach by serious- ly entertaining the possibility that the doors to knowledge are many,

13. Thanks to Ernest Sosa for suggesting this explanation, without necessarily endorsing it.

philosophers’ imprint – 15 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

Appendix A

I performed a binary logistic regression to assess the impact of Source and Content on the likelihood that participants would deny knowl- edge in Experiment 1. The model contained Source, Content, and the interaction of Source and Content as predictor variables, and knowl- edge denial as the outcome variable. The full model was statistically significant, χ2(3, N = 145) = 38, p < .000001 (see Table 1). It explained between 23.1% and 35.7% of the variance in knowledge denial, and it correctly classified 83% of cases. Neither Source nor Content of be- lief made a unique statistically significant contribution to the model. Table 2. Experiment 5: Logistic regression predicting knowl- There was a significant interaction between the Source and Content edge denial. Reference class for Source: Perception. Reference of belief, which recorded an odds ratio of 37.62. This indicates that by class for Content: Positive. changing the content of the protagonist’s belief from positive to nega- tive, the odds of denying knowledge increased over 37 times more for inferential belief than for perceptual belief. Appendix B

I performed a binary logistic regression to assess the impact of Source and Content on the likelihood that participants would deny knowl- edge in Experiment 5. The model contained Source, Content, and the interaction of Source and Content as predictor variables, and knowl- edge denial as the outcome variable. The full model was statistically significant, χ2(3, N = 160) = 23.61, p < .0001 (see Table 2). It explained between 13.7% and 20.8% of the variance in knowledge denial, and it correctly classified 77% of cases. Neither Source nor Content of be- lief made a unique statistically significant contribution to the model. Table 1. Experiment 1: Logistic regression predicting There was a significant interaction between the Source and Content knowledge denial. Reference class for Source: Perception. of belief, which recorded an odds ratio of 7.5. This indicates that by Reference class for Content: Positive. changing the content of the protagonist’s belief from positive to nega- tive, the odds of denying knowledge increased 7.5 times more for in- ferential belief than for perceptual belief.

philosophers’ imprint – 16 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

References Sosa (Eds.), Contemporary Debates in Epistemology (2nd ed.). Oxford: Wiley-Blackwell. Alicke, M. D. (1992). Culpable causation. Journal of Personality and Social Hume, D. (1748). An Enquiry Concerning Human Understanding. E. Stein- Psychology, 63(3), 368–378. berg (Ed.). Indianapolis: Hackett, 1993. Alicke, M. D., & Rose, D. (2010). Culpable control or moral ? Ichikawa, J. J., & Steup, M. (2012). The analysis of knowledge. The Stan- Behavioral and Brain Sciences, 33(04), 330–331. ford Encyclopedia of Philosophy. Ed. E. N. Zalta. Retrieved from http:// Cohen, S. (1988). How to be a fallibilist. Philosophical Perspectives, 2, .stanford.edu/archives/fall2013/entries/knowledge-analysis/. 91–123. Lewis, D. (1996). Elusive knowledge. Australasian Journal of Philosophy, Cohen, S. (1999). Contextualism, skepticism, and the structure of rea- 74(4), 549–567. sons. Philosophical Perspectives, 13, 57–89. Moore, G. E. (1959). Philosophical Papers. New York: Allen and Unwin. Cohen, S. (2013). Contextualism defended. In M. Steup, J. Turri, & E. Nagel, J., San Juan, V., & Mar, R. A. (2013). Lay denial of knowledge Sosa (Eds.), Contemporary Debates in Epistemology (2nd ed.). Oxford: for justified true beliefs. , 129(3): 652–661. doi:10.1016/j. Wiley Blackwell. cognition.2013.02.008. DeRose, K. (1995). Solving the skeptical problem. The Philosophical Re- Nozick, R. (1981). Philosophical Explanations. Cambridge, Mass.: Har- view, 104(1), 1–52. vard University Press. DeRose, K. (2009). The Case for Contextualism: Knowledge, Skepticism, Pillow, B. H. (1989). Early understanding of perception as a source of and Context, v. 1. Oxford: Oxford University Press. knowledge. Journal of Experimental Child Psychology, 47(1), 116–129. Dretske, F. I. (1970). Epistemic operators. The Journal of Philosophy, Priest, S. (2007). The British Empiricists (2nd ed.). New York: Routledge. 67(24), 1007–1023. Schaffer, J. (2007). Closure, contrast, and answer. Philosophical Studies, Dretske, F. I. (1981). Knowledge and the Flow of Information. Cambridge, 133(2), 233–255. Mass.: MIT Press. Sellars, W. (1963). Science, Perception and Reality. Atascadero, Dretske, F. (2013). The case against closure. In M. Steup, J. Turri, & E. Calif.: Ridgeview. Sosa (Eds.), Contemporary Debates in Epistemology (2nd ed.). Oxford: Sgarbi, M. (2012). The Aristotelian Tradition and the Rise of British Em- Wiley-Blackwell. piricism: Logic and Epistemology in the British Isles (1570–1689). Dor- Feldman, R. (1999). Contextualism and skepticism. Philosophical Per- drecht: Springer. spectives, 13, 91–114. Sosa, E. (1999). How to defeat opposition to Moore. Philosophical Per- Friedman, O., & Turri, J. (2014). Is probabilistic evidence a source of spectives, 13, 141–153. knowledge? Cognitive Science, forthcoming. Sosa, E. (2004). Relevant alternatives, contextualism included. Philo- Fumerton, R. (1987). Nozick’s epistemology. In S. Luper-Foy (Ed.), The sophical Studies, 119(1), 35–65. Possibility of Knowledge. Lanham, Md.: Rowman & Littlefield. Stanley, J. (2005). Knowledge and Practical Interests. Oxford: Oxford Uni- Hawthorne, J. (2004). Knowledge and Lotteries. Oxford: Oxford Univer- versity Press. sity Press. Starmans, C., & Friedman, O. (2012). The folk conception of knowl- Hawthorne, J. (2013). The case for closure. In M. Steup, J. Turri, & E. edge. Cognition, 124(3), 272–283. Steup, M. (2005). Epistemology. The Stanford Encyclopedia of Philosophy.

philosophers’ imprint – 17 – vol. 15, no. 2 (january 2015) john turri An Open and Shut Case

Ed. E. N. Zalta. Retrieved from http://plato.stanford.edu/entries/ Vogel, J. (1990). Are there counterexamples to the closure principle? epistemology/. In M. D. Roth & G. Ross (Eds.), Doubting. Dordrecht: Kluwer Aca- Stine, G. C. (1976). Skepticism, relevant alternatives, and deductive demic Publishers. Reprinted in Epistemology: An anthology (2nd ed.). closure. Philosophical Studies, 29(4), 249–261. Ed. E. Sosa, J. Kim, J. Fantl & M. McGrath. Oxford: Wiley-Blackwell. Stroud, Barry. (1984). The Significance of . Oxford: Williamson, T. (2000). Knowledge and Its Limits. Oxford: Oxford Uni- Clarendon Press. Excerpts reprinted in Epistemology: An Anthology versity Press. (2nd ed.). Ed. E. Sosa, J. Kim, J. Fantl & M. McGrath. Oxford: Wiley- Blackwell, 2008. Turri, J. (2010a). Epistemic invariantism and speech act con- textualism. The Philosophical Review, 119(1), 77–95. doi:10.1215/00318108–2009–026. Turri, J. (2010b). Does perceiving entail knowing? Theoria, 76(3), 197– 206. doi:10.1111/j.1755–2567.2010.01065.x. Turri, J. (2012). Pyrrhonian skepticism meets speech-act theo- ry. International Journal for the Study of Skepticism, 2(2), 83–98. doi:10.1163/221057011X588037. Turri, J. (2013a). The test of truth: an experimental investigation of the norm of assertion. Cognition, 129(2), 279–291. doi:10.1016/j. cognition.2013.06.012. Turri, J. (2013b). A conspicuous art: putting Gettier to the test. Philoso- phers’ Imprint, 13(10), 1–16. Turri, J. (2014a). Linguistic intuitions in context: a defense of nonskep- tical pure invariantism. In A. R. Booth & D. P. Rowbottom (Eds.), Intuitions. Oxford: Oxford University Press. Turri, J. (2014b). Epistemology: A Guide. Malden, Mass.: Wiley-Blackwell. Turri, J. (2014c). Skeptical appeal: the source-content bias. Cognitive Science. doi:10.1111/cogs.12153. Turri, J. (in press). Knowledge judgments in “Gettier” cases. In J. Syts- ma & W. Buckwalter (Eds.), A companion to experimental philosophy. Malden, Mass.: Wiley-Blackwell. Turri, J., & Blouw, P. (2014). Excuse validation: a study in rule-breaking. Philosophical Studies. doi:10.1007/s11098–014–0322-z. Turri, J., Buckwalter, W., & Blouw, P. (2014). Knowledge and luck. Psy- chonomic Bulletin & Review. doi:10.3758/s13423–014–0683–5.

philosophers’ imprint – 18 – vol. 15, no. 2 (january 2015)